ContactCenters
V. 0.9.9.

Package umontreal.iro.lecuyer.stat.mperiods

Provides facilities for storing observations during a simulation per time period, for the common situation where time is partitioned into a finite number of intervals, and statistics have to be collected separately for the different intervals.

See:
          Description

Interface Summary
MeasureMatrix Represents a matrix of measures for a set of related values during successive simulation periods.
 

Class Summary
IntegralMeasureMatrix<M extends MeasureMatrix> Computes per-period values for a matrix of measures with a single period.
IntegralMeasureMatrixSW<M extends MeasureMatrix> This extends IntegralMeasureMatrix to add a sliding window for the records.
ListOfStatProbesMeasureMatrix Matrix of measures whose values are obtained using an list of statistical probes.
MatrixOfStatProbesMeasureMatrix Matrix of measures whose values are obtained using a matrix of statistical probes.
MeasureSet Represents a set of related measures computed using different measure matrices.
MeasureSet.MeasureInfo Contains information about a measure added to a measure set.
StatProbeMeasureMatrix Matrix of measures whose value is obtained using a statistical probe.
SumMatrix This matrix of measures can be used to compute sums of values, or the number of occurrences of events.
SumMatrixSW Extends SumMatrix to add a sliding window.
 

Package umontreal.iro.lecuyer.stat.mperiods Description

Provides facilities for storing observations during a simulation per time period, for the common situation where time is partitioned into a finite number of intervals, and statistics have to be collected separately for the different intervals. For example, we may want to collect statistics on the quality of service for each hour in a telephone call center.

Sometimes, a vector or a matrix of statistical probes is sufficient for performing this task, especially when simulating independent replications. Probes are used during an experiment to compute sums and the resulting sums or averages are collected in tallies to get samples; the sample size corresponds to the number of replications. However, the number of periods can sometimes be random.This happens when using the batch means method [15], if the number of batches can change adaptively. For the same reason, it can be necessary to regroup periods to save memory, when the total number of periods becomes too large; this is not supported by a matrix of statistical probes. Events can also be counted in previous batches in addition to the last one. As a result, in a general setting, a probe computing a sum and being reset at the end of each batch cannot be used.

When the number of periods is large, it can be necessary to get observations for a subset of these periods, e.g., the last ten periods. For example, this can be used by the dialer of a phone call center to determine the quality of service in the last ten minutes on which decisions can be based.

To address these problems, this package defines an interface for a matrix of measures. Each row of this matrix corresponds to a type of event, e.g., the type of a customer in a retail store. Each column corresponds to a period that can be any time interval such as half an hour, a complete day, etc.

When independent replications of the same simulation are performed, the finite horizon is often divided into periods. For each period, a vector of observations may be computed and stored. This results in a matrix of observations that can be added to a matrix of tallies at the end of each replication. The matrix can be obtained from the vectors of observations directly, or some vectors can be regrouped.

When simulating for an infinite horizon, a single vector of observations may be obtained. However, to get an estimate on the variance, for computing confidence intervals, the simulation time is divided into intervals called batches. For each batch, a vector of observations is obtained. If the simulation length is constant, or if the number of batches is allowed to be random, it is sufficient to compute one vector at a time and collect it as observations at the end of each batch. However, when the number of batches is required to be constant while the simulation length is random, it is necessary to keep all the vectors of observations to regroup them later. Vectors of observations, or groups of vectors, are collected during or at the end of the simulation.

Depending on the type of experiment, matrices of measures can be added directly to tallies as matrices of observations, some columns can be regrouped, or each column (or group of consecutive columns) might be collected in tallies separately. This package provides a mechanism to implement simulation events collecting independently of the way the experiment is performed. The MeasureMatrix interface represents a matrix of measures. Usually, an implementation of this interface counts the number of occurrences of an event, the sum of values, or some integrals. The SumMatrix class implements this interface for computing a matrix of sums. A subclass, SumMatrixSW, provides a sliding window permitting the observations in a subset of the periods to be stored. For integrals, the simulator must provide a custom implementation of MeasureMatrix which computes a function relative to the simulation time, from time 0 to the current simulation time. The IntegralMeasureMatrix class can be used to get the value of the integral for time intervals by storing the value of the integral for user-defined times.


ContactCenters
V. 0.9.9.

To submit a bug or ask questions, send an e-mail to Richard Simard.